EVIEWS中异方差性检验及补救 | 您所在的位置:网站首页 › eviews用图示法检验异方差 › EVIEWS中异方差性检验及补救 |
目的:1、正确使用EVIEWS 2、会使用OLS和WLS,Goldfeld-Quandt检验 3、能根据计算结果进行异方差分析和出现异方差性后的补救。 3、数据为demo data1
实例:某市人均储蓄与人均收入的关系分析(异方差性检验及补救)
根据某市1978-1998年人均储蓄与人均收入的数据资料(见下表),其中X为人均收入(元),Y为人均储蓄(元),经分析人均储蓄受人均收入的线性影响,可建立一元线性回归模型进行分析。
obs X Y 1978 590.2000 107.0000 1979 664.9400 123.0000 1980 809.5000 159.0000 1981 875.5400 189.0000 1982 991.2500 233.0000 1983 1109.950 312.0000 1984 1357.870 401.0000 1985 1682.800 522.0000 1986 1890.580 664.0000 1987 2098.250 871.0000 1988 2499.580 1033.000 1989 2827.730 1589.000 1990 3084.170 2209.000 1991 3462.710 2878.000 1992 3932.520 3722.000 1993 5150.790 5350.000 1994 7153.350 8080.000 1995 9076.850 11758.00 1996 10448.21 15839.00 1997 11575.48 18196.00 1998 12500.84 20954.00
1、用OLS估计法估计参数 设模型为: 运行EVIEWS软件,并输入数据,得计算结果如下:
Dependent Variable: Y Method: Least Squares Date: 10/11/05 Time: 23:10 Sample: 1978 1998 Included observations: 21 Variable Coefficient Std. Error t-Statistic Prob. C -2185.998 339.9020 -6.431262 0.0000 X 1.684158 0.062166 27.09150 0.0000 R-squared 0.974766 Mean dependent var 4533.238 Adjusted R-squared 0.973438 S.D. dependent var 6535.103 S.E. of regression 1065.086 Akaike info criterion 16.86989 Sum squared resid 21553736 Schwarz criterion 16.96937 Log likelihood -175.1338 F-statistic 733.9495 Durbin-Watson stat 0.293421 Prob(F-statistic) 0.000000
2、异方差检验 (1)Goldfeld-Quandt检验 在Procs菜单项选Sort series项,出现排序对话框,输入X,OK。 在Sample菜单里,将时间定义为1978-1985,用OLS方法计算得如下结果: Y = -145.441495 + 0.3971185479*X (-8.730234) (25.42693) R-squared=0.990805 Sum squared resid1=15.12284
Dependent Variable: Y Method: Least Squares Date: 10/11/05 Time: 23:25 Sample: 1978 1985 Included observations: 8 Variable Coefficient Std. Error t-Statistic Prob. C -145.4415 16.65952 -8.730234 0.0001 X 0.397119 0.015618 25.42693 0.0000 R-squared 0.990805 Mean dependent var 255.7500 Adjusted R-squared 0.989273 S.D. dependent var 146.0105 S.E. of regression 15.12284 Akaike info criterion 8.482607 Sum squared resid 1372.202 Schwarz criterion 8.502468 Log likelihood -31.93043 F-statistic 646.5287 Durbin-Watson stat 1.335534 Prob(F-statistic) 0.000000
在Sample菜单里,将时间定义为1991-1998,用OLS方法计算得如下结果: Y = -4602.367144 + 1.952519317*X (-5.065962) (18.40942) R-squared=0.982604 Sum squared resid2=5811189.
Dependent Variable: Y
Method: Least Squares
Date: 10/11/05 Time: 23:29
Sample: 1991 1998
Included observations: 8
Variable Coefficient Std. Error t-Statistic Prob. C -4602.367 908.4882 -5.065962 0.0023 X 1.952519 0.106061 18.40942 0.0000 R-squared 0.982604 Mean dependent var 10847.12 Adjusted R-squared 0.979705 S.D. dependent var 6908.102 S.E. of regression 984.1400 Akaike info criterion 16.83373 Sum squared resid 5811189. Schwarz criterion 16.85359 Log likelihood -65.33492 F-statistic 338.9068 Durbin-Watson stat 0.837367 Prob(F-statistic) 0.000002
求F统计量: 3、异方差的修正 (1)WLS估计法。 首先生成权函数 Y = -2262.639946 + 1.566910934*X Dependent Variable: Y Method: Least Squares Date: 10/12/05 Time: 08:07 Sample: 1978 1998 Included observations: 21 Weighting series: W Variable Coefficient Std. Error t-Statistic Prob. C -2262.640 131.2507 -17.23907 0.0000 X 1.566911 0.057637 27.18590 0.0000 Weighted Statistics
R-squared 0.961501 Mean dependent var 2183.201 Adjusted R-squared 0.959475 S.D. dependent var 2104.209 S.E. of regression 423.5951 Akaike info criterion 15.02583 Sum squared resid 3409224. Schwarz criterion 15.12530 Log likelihood -155.7712 F-statistic 474.5211 Durbin-Watson stat 0.354490 Prob(F-statistic) 0.000000 Unweighted Statistics
R-squared 0.962755 Mean dependent var 4533.238 Adjusted R-squared 0.960794 S.D. dependent var 6535.103
S.E. of regression 1293.978 Sum squared resid 31813191
Durbin-Watson stat 0.224165
(2)对数变换法。 用GENR生成LY和LX序列,用OLS方法求LY 对LX的回归,结果如下: LY = -6.839135503 + 1.787148637*LX Dependent Variable: LY Method: Least Squares Date: 10/12/05 Time: 00:05 Sample: 1978 1998 Included observations: 21 Variable Coefficient Std. Error t-Statistic Prob. C -6.839136 0.237565 -28.78845 0.0000 LX 1.787149 0.030033 59.50680 0.0000 R-squared 0.994663 Mean dependent var 7.195082 Adjusted R-squared 0.994382 S.D. dependent var 1.746173 S.E. of regression 0.130880 Akaike info criterion -1.138677 Sum squared resid 0.325463 Schwarz criterion -1.039199 Log likelihood 13.95611 F-statistic 3541.059 Durbin-Watson stat 0.642916 Prob(F-statistic) 0.000000
比较方法(1)和(2),可以看出X与Y在对数线性回归下拟合效果较好。原因是Y的曲线呈对数型图形有关。 |
CopyRight 2018-2019 实验室设备网 版权所有 |